Subspace learning with enriched databases using symmetry

Abstract

Principal Component Analysis and Linear Discriminant Analysis are of the most known subspace learning techniques. In this paper, a way for training set enrichment is proposed in order to improve the performance of the subspace learning techniques by exploiting the a-priori knowledge that many types of data are symmetric. Experiments on artificial, facial expression recognition, face recognition and object categorization databases denote the robustness of the proposed approach.